Stylus input

Labello handles a primary pen tip as an annotation pointer. Drag to create a box, drag a selected box to move it, and drag its handles to resize it. Tap to place a keypoint and drag a placed keypoint to move it. Pan mode and the configured pan modifier still take priority. Pressure and tilt do not change annotation geometry.

Support and conformance

The browser adapter accepts PointerEvent streams with pointerType = "pen", isPrimary = true, and primary-tip contact. It reads pointermove directly; compatibility mouse movement is not required. Pen compatibility mouse events and touch events whose touchType is stylus do not enter the canvas twice. The browser canvas disables native page pan/zoom during direct manipulation.

OS, browser, and event source	Evidence and support boundary
Ubuntu 24.04, Chromium 149.0.7827.55, CDP-emulated primary pen	Browser-generated pen events reach the real WASM client and API. Box create/move/resize, keypoint place/move, and sequential pen/mouse/touch use are checked. Overlapping scripted pen and trusted touch streams additionally check both contact orders. This is an emulated event contract, not a physical pen certification.
Ubuntu 24.04, Playwright WebKit 26.5, scripted primary pen	Checks the same production client with explicit DOM events, including movement without compatibility mouse events. This is desktop WebKit, not iPadOS Safari.
iPadOS Safari with Apple Pencil or another pen reporting pointerType = "pen"	Priority physical target. Device model, iPadOS/browser version, rapid repeated contact, and mixed-input results must be recorded with the procedure below before claiming verified support. Currently unverified.
iPadOS Firefox with the same pen	Priority physical target and a separate application test. Firefox for iOS uses WebKit, but passing desktop WebKit or Safari does not establish Firefox application behavior. Currently unverified.
Linux with a physical pen tablet; Android Chrome or Firefox with an active stylus	Candidate combinations. Chromium emulation does not establish OS/driver behavior. Currently unverified.
Passive capacitive stylus reported only as a finger, or legacy touch-only pen delivery	No distinct pen support. These follow ordinary touch behavior and cannot obtain pen-specific conflict handling.

Mozilla documents Firefox's iOS engine in its Firefox for iOS architecture. WebKit added generic mouse, touch, and stylus Pointer Events in Safari 13. Neither source establishes Labello's physical-device compatibility.

The original broad US-03 stylus requirement remains partially satisfied. The input adapter and emulated annotation contract have regression coverage; iPadOS Safari/Firefox physical acceptance remains pending. No physical-device combination is certified by the automated check.

Unsupported features include pressure-sensitive drawing, tilt-dependent tools, eraser/barrel-button tools, Pencil hover previews beyond ordinary pointer hover, Scribble, and OS palm-rejection guarantees. Simultaneous pen and mouse gestures remain unsupported. Two-finger pan/pinch and pen annotation use separate input streams and can overlap in either contact order. Navigation changes the image transform while the pen edit continues in image coordinates. Finger movement and release must not commit or cancel the pen edit. Holding the pen still while fingers are down must not turn it into a touch long-press or discard the canvas's drag ownership.

After the first pen hover or contact in a browser session, fingers on the image canvas navigate with two fingers and do not create annotations. Browsers do not provide a reliable connected-stylus inventory, so this policy starts with the first observed pen event, not hardware connection. Reloading starts a fresh session. Touch-only sessions retain their ordinary annotation behavior. Finger presses on controls outside the image still work. Pen drags over scrollable controls do not drag-scroll their contents; fingers, scrollbars, and wheel input retain their scrolling behavior.

Cancelled or lost pen capture discards the edit preview. Escape also cancels an annotation drag. These input rules do not certify hardware palm rejection.

Automated browser procedure

The focused check runs the production release WASM distribution against a new loopback API server. It creates its own synthetic image and dataset through the API, authenticates through local development login, and verifies saved annotation versions and geometry. It never connects to an existing server or browser profile. Each invocation uses fresh temporary data and cleans up its server and browser on success, failure, timeout, or interruption.

Prerequisites are CPython 3.11 or newer, the repository's locked build tools, and Playwright's browser dependencies. Install into a private environment:

python3.11 -m venv /tmp/labello-stylus-env
/tmp/labello-stylus-env/bin/pip install -r apps/labello-wasm/tests/stylus-requirements.txt
/tmp/labello-stylus-env/bin/playwright install chromium webkit
cargo build --locked -p labello-server
cd apps/labello-wasm
trunk build --release --locked

Return to the repository root. Run both annotation types through Chromium's trusted emulated pen stream, then through the explicit DOM regression streams:

for kind in bounding_box skeleton; do
  /tmp/labello-stylus-env/bin/python apps/labello-wasm/tests/stylus_input.py --kind "$kind"
  /tmp/labello-stylus-env/bin/python apps/labello-wasm/tests/stylus_input.py --kind "$kind" --events pointer-only
  /tmp/labello-stylus-env/bin/python apps/labello-wasm/tests/stylus_input.py --kind "$kind" --events dom-compat
  /tmp/labello-stylus-env/bin/python apps/labello-wasm/tests/stylus_input.py --kind "$kind" --browser webkit --events pointer-only
done

pointer-only supplies no compatibility mouse movement. dom-compat adds redundant stylus touch and mouse events. These explicit DOM streams are synthetic and alternate touch-before-pointer and pointer-before-touch delivery. They are untrusted, unlike Chromium's CDP-generated events. They test adapter behavior, not the events an iPad actually produces. Trusted touch injection and Chromium's canvas accessibility node are checked only in Chromium; the report says whether the touch check ran. WebKit does not provide the CDP injection used by that check. The WebKit check accepts only pointer-only: this desktop build also lacks a usable Touch constructor for the scripted compatibility stream. Pointer-cancellation and capture-loss probes use scripted events in every mode. Chromium also runs overlapping scripted pen and trusted CDP finger gestures in both contact orders, verifies the saved edit through the changed view transform, and checks that a single finger cannot annotate after pen use. A short inspector check verifies that a pen drag does not scroll controls while a wheel still does. This is separate from the trusted CDP pen checks; it does not prove physical simultaneous delivery.

Repeat relevant cases with --width 600 --height 800 --dpr 2, --width 390 --height 844 --dpr 3, or the other applicable sizes in the UI matrix. DPR and viewport emulation are not actual browser zoom or a physical tablet. The automated procedure runs at 100% zoom; use the physical procedure for actual zoom and OS behavior.

The check fails within two minutes and prints a bounded JSON outcome with browser version, viewport, DPR, event source, and checks actually performed. It retains no screenshots, geometry, HTTP payloads, cookies, or browser storage. It uses screenshots only in memory to locate the generated flat-color fixture and check drag previews. Two colored reference marks locate the rendered image transform for concurrent edits even when zoom clips the image edges. Do not extend it to arbitrary datasets or broad browser traces. LABELLO_TEST_WEBKIT_EXECUTABLE can select a locally provisioned Playwright WebKit executable when browser libraries are installed outside system paths.

Use --negative-control --events pointer-only to suppress pen presses. The command must exit nonzero. The original missing movement path is reproduced by running --events pointer-only against the pre-adapter distribution: the final release position can be correct, but the moving preview assertion fails.

Run the same script with --touch-only in Chromium to verify that a finger-only session can still create and save a box before any pen event. This checks normal touch release ordering separately from pen and touch cancellation.

This focused manual browser check supplements canonical verification. It is not the general browser workflow suite and is not automatically run by Testing.

Physical iPad procedure

Use a disposable dataset on an authenticated test deployment of the exact candidate revision. Do not enable loopback-only local administrator login on a network-facing deployment to make it reachable from the iPad.

Record the candidate commit, iPad model, iPadOS version, stylus model, Safari version, Firefox application version, orientation, viewport, DPR, browser zoom, and relevant Pencil/Scribble settings. Test Safari and Firefox independently. Use an artificial image and separate box and skeleton workflows.

1. At default zoom, create one box with the pen. Confirm its preview follows the tip throughout the drag, then wait for Saved. Move it and resize from each handle. Reload and confirm one object with the intended geometry.
2. Place every keypoint of a skeleton, then move placed keypoints. Include rapid successive contacts and contacts without hover. Confirm exactly one placement or edit per intended action, no accidental Fit, and no image pan or zoom.
3. Lift the pen, use a mouse or trackpad if available, then use the pen again. Separately, lift the pen, perform a two-finger pan/pinch, lift both fingers, Fit, and annotate again. Check saved object and keypoint counts after each.
4. While a pen box/keypoint edit is active, pan/pinch with two fingers. Confirm navigation works, the edit continues under the pen, and releasing fingers does not commit or cancel it. Repeat with fingers touching first. Check saved geometry after pen release. Then drag one finger on blank image space after lifting the pen; it must not add annotations. Record palm contacts separately; do not infer hardware palm rejection from the scripted test.
5. Start a drag and interrupt it by switching apps or changing orientation. Confirm no partial edit is saved and the next gesture works. Where a keyboard is attached, verify Escape cancellation and the configured pan modifier.
6. Test the surrounding controls with the pen, including opening/dismissing drawers, changing the workflow, Fit, Undo/Redo, and opening a browser file picker in the disposable admin workflow. Browser user activation must survive the pen adapter.
7. Repeat in landscape, portrait, Split View, and actual 200% browser zoom. Verify complete control reachability and that the annotation follows the tip.

Record each step as passed, failed, or unavailable with a bounded symptom. Follow the artifact redaction rules. A table of outcomes is sufficient; do not publish image content, annotation geometry, credentials, request bodies, or broad recordings of the runtime dataset. Promote only combinations that pass all applicable steps, and retain explicit limitations for unavailable peripherals or untested settings.